Documente Academic
Documente Profesional
Documente Cultură
Title
Conceptual design of a compact positron tomograph for prostate imaging
Permalink
https://escholarship.org/uc/item/39w571wf
Authors
Huber, J.S.
Derenzo, S.E.
Qi, J.
et al.
Publication Date
2000-11-04
Peer reviewed
Prostate
(b)
Upper Detector “Ring”
Adjustment Direction
Septa −100
−50
50
Lead Shielding
Fig. 3: Cut-out side view of the proposed positron tomograph, showing the 100
shielding and inter-plane septa.
150
−200 −100 0 100 200
penetration effects, but these FOV regions are less important.
Both detector banks are tilted to image the prostate while Fig. 5: Statistical noise-free reconstructed image of five point sources,
demonstrating the local impulse response function and sampling
minimizing attenuation (i.e., above the buttocks and below
sufficiency. Scale in millimeters.
the stomach), but the gantry can also allow zero tilt for thin
patients. The patient is not fully encircled in 2D, which will coincidences and from photons that Compton scatter in the
be discussed further in Section III. patient. Lead shields are also used on the ends to reduce
Standard block detectors with thick gamma ray detectors activity from outside the field of view.
(i.e., 3 attenuation lengths) are needed for good detection
efficiency. Narrow detector elements (i.e., 4.5 x 4.5 x 30 mm III. EXPECTED PERFORMANCE
BGO crystals) are desired to achieve good spatial resolution.
The 511 keV photons of primary interest originate from the A. Simulation
prostate which is roughly a small (e.g., 3 cm diameter) Our camera geometry provides a compact design (i.e., 44
spherical source with a known central location at a reasonable cm minor axis between opposing detector faces) that adjusts
distance from detector modules, so expensive detector vertically to fit the patient. However, a patient of average size
modules with depth of interaction capability are not will not be fully encircled by the detector rings, as mentioned
necessary. A two ring camera is more practical for patient in Section II. Fig. 4 shows the resulting irregular and
positioning. Therefore, our proposed camera design attempts incomplete sampling due to the side gaps. Similar irregular
to maximize performance while minimizing cost, in order to and incomplete sampling patterns have been dealt with in our
produce a PET camera with nearly twice the detection positron emission mammography camera [21]. Despite this
efficiency as a conventional 2D PET camera with incomplete sampling, we are able to reconstruct nearly artifact
approximately one-quarter the detector costs. free images in the region of interest by using an iterative
The camera design also includes shielding, as shown in reconstruction as described below. This iterative
Fig. 3. Inter-plane septa extend 5 cm beyond the scintillator reconstruction algorithm can easily deal with the missing
crystals to reduce background events from random data of the system and can also partially recover the
100 resolution loss (due to the crystal penetration effect) in the
80 off-center FOV region. The gap size will increase for larger
patients and this may impact the image quality, but we have
60
not yet explored the effects of variations in gap size.
40 Our camera design includes inter-plane septa, so the data
will be acquired in 2D. We use Monte Carlo data to simulate
Angle (Degree)
20
the camera performance. Reconstructions are performed with
0 an iterative maximum likelihood algorithm [22]. Self-
−20 attenuation is simulated assuming a constant attenuation
coefficient of 0.0095/mm inside the body, then attenuation is
−40
modeled in the reconstruction. Coincidences between detector
−60 modules within the same bank are allowed.
−80
Fig. 5 shows an image of five point sources that are
reconstructed without statistical noise to demonstrate how
−100 irregular sampling affects the local response function. All
−300 −200 −100 0 100 200 300
Radial Distance (mm) point sources are clearly visible. We observe increased radial
Fig. 4: Incomplete sampling for the field of view due to side gaps between blurring near the outer surface of the patient as expected due
detector banks. Coincidences between detector modules within the same to penetration effects, but observe good resolution in the
bank and imaging plane are allowed, resulting in the ~triangular sampling
regions on the left and right. The field of view of the proposed camera is region near the center (i.e., near the prostate). In short, we are
an elliptical region, so the only missing samples are the “X” region in the optimizing resolution at the center near the prostate at the
center of the sinogram. price of resolution at the edges, which implies that distant
4
Accepted to IEEE Transactions on Nuclear Science LBNL-47068
metastases will suffer from some resolution degradation. Data shows the anticipated spatial resolution obtained in this
are generated by forward projection of 256x256 2 mm x 2 manner as a function of distance from the center of the field
mm phantom voxels to simulate real data, and reconstruction of view (along both the x- and y- axes). Notice that little
is done with 128x128 4.51 mm x 4.51 mm pixels, thus degradation due to penetration artifacts is seen for objects
causing some point source spread among adjacent pixels. within 5 cm of the camera center, and that 6 mm fwhm or
As we are planning to use iterative methods to reconstruct better resolution is obtained within 8 cm of the camera
images, the anticipated spatial resolution will depend on the center. This region is easily large enough to contain the entire
geometrical parameters used for reconstruction (e.g., the prostate.
voxel size and detector model) as well as the number of We also use Monte Carlo data to simulate the performance
events. Thus, it is difficult to derive meaningful estimates of of our proposed camera for imaging extended sources.
the expected spatial resolution from the test image in Fig. 5. Extended “prostate” sources are reconstructed with statistical
Therefore, we estimate the spatial resolution analytically with noise representing 745 kcounts using a preliminary iterative
the help of a simplifying assumption. That assumption is maximum likelihood algorithm. We assume a 2.5 cm
that if two tomographs employ the same scintillation diameter spherical “prostate” (see fig. 7(a)) with a tumor to
material, the contribution of penetration to the spatial background ratio of 2:1, a 30% scatter fraction and 20%
resolution for a source at any position within either camera randoms fraction. The images are post-filtered with a
depends only on the angle θ that the incoming gamma ray Gaussian function having a standard deviation of 1 pixel. The
makes to the axis of the detector crystals (see Fig. 6). Given spherical "prostate" is divided into quadrants. Fig. 7 (b)
this assumption, we can use measured spatial resolutions [23] shows a reconstructed image with a tumor in the upper left
to estimate the resolution for the proposed camera. Fig. 6(b) quadrant of the spherical prostate. Fig. 7 (c) shows a
reconstructed image with a tumor in the right half of the
(a) prostate. Fig 7 (d) shows a reconstructed image with a tumor
in the upper right and lower left quadrants. These "prostate"
tumors are clearly visible in all images, demonstrating that
we can reconstruct nearly artifact free images near the prostate
despite incomplete sampling and that the scanner can
θ differentiate partial and whole prostate tumors. Our
y preliminary rate calculations (described below) indicate that
x 745 kcounts per imaging plane will be achievable with a 6-
minute scan after injecting a clinically acceptable (10 mCi)
patient dose of [11C]choline or [18F]fluorocholine. A relatively
short value of six minutes is chosen to demonstrate that good
images can be obtained even if the bladder fills rapidly. All
published [11C]choline or [18F]fluorocholine studies done to
(b) date have used three minutes or longer (up to 30 minutes)
16 imaging times.
Radial Resolution (mm fwhm)